CyrusOne

Phase A — Understand the business

Lens 1 · Company Overview

CyrusOne is a wholesale / hyperscale data-center developer and operator — a landlord of powered, cooled, physically secure computing space leased to the largest technology companies on earth. It is one of the "big two independents" of the hyperscale-colo world alongside Vantage, and one tier removed from the retail-interconnection incumbents Equinix and Digital Realty. Founded in Houston in 2001, IPO'd 2013, taken private 2022.

What it actually sells. Long-duration triple-net-style leases on turn-key powered shells — the customer (a hyperscaler) brings its own servers/GPUs; CyrusOne delivers the building, the power feed, the cooling, redundancy, and physical security, and collects rent per kilowatt per month for 10–15 years. This is infrastructure, not cloud: CyrusOne does not touch the compute. The revenue is contracted, recurring, and inflation-escalated — closer to a toll road than a tech company. This is the reason KKR/GIP (infrastructure investors, not tech investors) bought it.

Scale today ``:

• ~55+ data centers across North America, Europe, and Asia
• ~1,000 MW of operating power capacity across ~5 million sq ft of colocation space
• ~1,000 customers, including ~200 Fortune 1000 companies
• Ranked #8 on Data Centre Magazine's 2025 Top-100 global operators list ``

Contract structure — the critical feature. Hyperscale data-center leases are typically one tenant per building, long-dated (10–15 yr), with contractual annual escalators and, increasingly, power pass-throughs. This gives extraordinary revenue visibility but embeds single-tenant concentration by design — the moat and the risk are the same fact (see Lens 13). Churn is structurally low: a hyperscaler that has installed tens of thousands of GPUs in a building does not move them.

Customers, suppliers, competitors (detailed in Lenses 2–3): customers = Microsoft, Meta (Facebook), AWS, and increasingly AI/neocloud tenants; suppliers = power utilities/IPPs (Calpine, Constellation, KEPCO, E.ON), electrical-gear OEMs (Vertiv, Eaton, ABB, Schneider), and EPC/GC builders; competitors = Vantage, QTS (Blackstone), Aligned, STACK, Compass, DataBank, plus Equinix/Digital Realty at the margins.

Lens 2 · Supply Chain — name every stakeholder or it didn't happen

The value chain runs land + power → shell + fit-out → leased compute space → hyperscaler → end AI/cloud workload. Named stakeholders along CyrusOne's actual chain ``:

Upstream — power (the binding constraint):

• Calpine — 400 MW powered-land agreement at the Thad Hill Energy Center, Bosque County, TX (originally 190 MW, expanded +210 MW); plus a further 380 MW Phase-2 exclusive at Freestone ``.
• Constellation Energy — 380 MW agreement to serve a new facility adjacent to the Freestone Energy Center, Fairfield, TX ``.
• E.ON — on-site gas-fired generation (61 MW) at FRA7, Frankfurt, to bypass grid constraints; a strategic European partnership ``.
• Kansai Electric Power (KEPCO) — 50:50 JV for Japan (≥¥1trn / ~$7B over a decade, targeting ~900 MW / 600 MW IT) ``.
• Combined, CyrusOne has assembled ~1.16 GW of committed + scalable powered-land capacity across just two Texas campuses via Calpine/Constellation `` — a deliberate answer to the #1 industry bottleneck.

Midstream — construction & critical equipment:

• EPC / general contractors (e.g. Moss Utilities on the Lake Whitney/Whitney campus ``) — sitework, shell, MEP.
• Electrical & thermal OEMs — the same names that show up as public peers in the index: Vertiv, Eaton, ABB, Schneider, Generac/backup gensets, Trane/liquid-cooling. These are the long-lead-time chokepoints (switchgear, transformers, chillers) that gate delivery industry-wide.

Downstream — tenants → workloads:

• Hyperscalers (Microsoft, Meta, AWS) and increasingly AI/neocloud operators. The end workload is now AI training/inference, which is what has pulled data-center demand vertical since 2023.

Chokepoints / single-source dependencies:

1. Power interconnection — 3–5 yr allocation delays in top markets (Northern Virginia) ``. CyrusOne's response (behind-the-meter gas via Calpine/E.ON, powered-land JVs) is its most important strategic differentiator.
2. Long-lead electrical gear — transformers/switchgear industry-wide 1–2 yr lead times.
3. The building itself is single-tenant — see Lens 13.

Lens 3 · Competitive Advantages (moats)

CyrusOne's moat is real but not proprietary — it is a scale-and-speed-and-power moat, the kind that protects a top-5 incumbent but does not confer pricing monopoly.

Durable advantages:

• Powered land bank. In a market where power, not capital, is the scarce input, CyrusOne's early, exclusive powered-land deals (Calpine, Constellation, E.ON, KEPCO) are the single most defensible asset — ~1.16 GW pre-secured in Texas alone. You cannot buy your way past a 3–5 yr interconnection queue; you have to have signed years ago. This is genuine.
• Sponsor capital + investment-grade-style balance-sheet engineering. KKR/GIP backing plus the CMBS/ABS securitization machine (Lens 5) gives CyrusOne access to ~$12B/yr of development capital at scale most privates cannot match. Capital is a moat when each campus costs $1–2B.
• Hyperscaler relationships & delivery track record. ~1,000 customers, deep repeat business with Microsoft/Meta. Hyperscalers concentrate spend on a short list of landlords who can deliver at gigawatt scale on schedule — CyrusOne is on that list.
• Switching costs (per building). Once a tenant installs compute, the lease is effectively sticky for its full term.

Bargaining power — the honest read: CyrusOne has strong power over its own suppliers of construction (it's a huge, repeat buyer) but weak-to-balanced power over its hyperscaler customers, who are far larger than it, run multi-vendor procurement, and can (and do) self-build. The current seller's market (1.4% vacancy, +20–35% YoY rents — Lens 5/11) flips pricing power to landlords temporarily and cyclically; that is a market condition, not a CyrusOne moat, and it will normalize as supply catches up. Do not confuse the 2025 rent spike with durable competitive advantage.

Weakest moat leg: there is no network effect and no meaningful IP. A wholesale powered shell is a commodity differentiated by location, power, and delivery speed. Equinix has an interconnection/ecosystem moat CyrusOne structurally lacks — CyrusOne chose the wholesale lane, which is higher-volume, lower-margin, and more contestable.

Lens 4 · Segments

No audited segment disclosure exists post-2022 (segments.csv empty; private). Structural segmentation from web sources ``:

• By product: overwhelmingly hyperscale wholesale (large-footprint, single-tenant, wholesale pricing), with a legacy enterprise/colocation tail. In its last public years, hyperscale was ~70% of new bookings ``, and the mix has only tilted further hyperscale/AI since.
• By geography: US-dominant (Texas — DFW/San Antonio/Houston — plus Northern Virginia the core markets), a substantial European platform (Frankfurt, London, Dublin, Amsterdam, Milan, Madrid), and a nascent Asia footprint (Japan via KEPCO JV — Osaka OSK1, 48 MW, first phase ~Q1 2028) ``.

Trend & cause: the growth vector is unambiguously accelerating and shifting to AI-scale hyperscale — 400 MW single deals (Thad Hill), gigawatt powered-land campuses, international greenfield (Milan MIL1/MIL2, Japan). The enterprise-colo legacy is a shrinking share of a fast-growing pie. Caveat: precise segment revenue is n/a — private, not disclosed.

---

Phase B — Measure performance

Lens 5 · Earnings Result → (Overlay) Funding & Valuation Trajectory + Traction

No quarterly print exists (private). Under the +private overlay this lens becomes funding, valuation, and financial-engineering trajectory, which for CyrusOne is the story — it is run as a capital-markets machine.

Last audited baseline (FY2021, final public year) ``:

• Revenue $1.21B, +16.7% YoY
• Net income $25.3M (REIT — heavy D&A depresses GAAP net income; the relevant metric is FFO)
• Adjusted EBITDA ~$580M, +8.0% YoY
• Normalized FFO ~$494M (sum of quarters: $120.2 + $123.1 + $127.2 + $123.9M) ``
• Record US lease signings driven by Meta (Facebook) ``

Take-private valuation: $15B enterprise value, $90.50/share all-cash (a modest ~2.5% premium to the last close but a large premium to pre-rumor levels), closed 25 Mar 2022 — the largest data-center M&A deal in history at the time ``.

Estimated current scale. One aggregator (companiesmarketcap, using stale CONE data) cites TTM revenue ~$1.20B — **treat as unreliable/stale**; it almost certainly understates the private entity, which has roughly doubled its footprint and operates in the tightest rental market on record. A defensible: with ~1,000 MW operating (roughly 2× the ~500 MW-equivalent of 2021) and materially higher per-kW rents, current run-rate revenue is plausibly ~$2.0–2.5B ``. Not sourced to a hard figure — flagged as estimate.

The financing machine (the real "earnings" of a private infra platform) ``:

• 2024: ~$12B raised in a single year — pioneering the data-center CMBS/ABS market:
• $7.9B warehouse credit facility (Morgan Stanley, TD Securities, KKR Capital Markets; closed Jul 2024)
• $1.8B revolving credit facility (Wells Fargo; May 2024)
• $1.175B ABS issuance
• $687.1M SASB CMBS on DFW1 — the first co-location/enterprise SASB securitization since 2008 ``
• Conflict to surface: CyrusOne's own PR headlined "$9.7B in new debt" while Data Center Frontier totals "nearly $12B" — the gap is the warehouse-facility commitment vs. drawn/term amounts. Both are cited; do not merge into one figure.
• 2025: $575M ABS (13 DCs in N. Virginia/San Antonio/Houston, $4.4B appraised collateral — "highest comparable" data-center securitization) ; a **$701M CMBS**; a **~$1.05B refinancing** of DFW3/DFW4 in Allen .
• The red flag (Dec 2025): Goldman Sachs paused a planned ~$1.3B mortgage-bond sale for CyrusOne immediately after the CME outage (Lens 8/10) — the first visible crack in the securitization flywheel ``.

Read: CyrusOne is a levered, securitization-funded infrastructure roll-up. Its "quarterly result" is really "can it keep issuing asset-backed paper at attractive spreads to fund a multi-gigawatt build?" — and in Dec 2025 that machine hit its first pothole.

Lens 6 · Sentiment Trend → (Overlay) Founder/Exec Signal & Narrative

No earnings calls (private). Overlay: management's public narrative via interviews/PR. Consistent themes across 2024–2026 ``:

• "Power first." CEO Eric Schwartz frames the business around power, grid connectivity, and site infrastructure — every major 2025–2026 announcement leads with a utility partner (Calpine, Constellation, E.ON, KEPCO). The narrative has shifted decisively from "space" to "power" — the correct read of where the bottleneck moved.
• "Scale and speed for AI." Constant emphasis on gigawatt-scale delivery, high-density/AI-ready design (Intelliscale™), and accelerated timelines.
• International expansion (Italy, Japan) as the growth frontier.
• What they stopped saying: the old REIT vocabulary (FFO/AFFO, dividend, same-store) is gone — consistent with private, growth-capex-funded ownership where cash is reinvested, not distributed.
• Post-CME (late 2025): the narrative pivoted to operational-integrity damage control — attributing the outage to human error (staff/contractors failing to drain cooling towers before a freeze) ``. Reputationally defensive, not offensive.

Lens 7 · Comps → (Overlay) Cap Table, Sponsors & Peer Marks

Ownership / cap table ``:

• Owners: KKR + Global Infrastructure Partners (GIP) — 100% since Mar 2022. (GIP was itself acquired by BlackRock in 2024, so BlackRock now sits behind the GIP stake — a notable strategic backer.) Syndicate quality: top-tier infrastructure sponsors, exactly the crossover-caliber capital the overlay flags as an IPO-proximity / durability tell.
• Hold clock: ~4+ years into ownership as of mid-2026 — squarely in the PE exit window (sponsor-to-sponsor sale, infra-LP stake sale, or re-IPO all plausible) ``. No confirmed recap or stake sale found — a searched-for "$50B / Blue Owl" transaction could not be verified and is treated as n/a.

Peer table — private marks are n/a where undisclosed; public-peer multiples are `` where sourced and n/a otherwise (provenance discipline: no fabricated multiples).

Read on relative position: on tracked wholesale-capacity share, Vantage (~33%) and QTS (~38%) now out-scale CyrusOne in the specific hyperscale-development segment `` — CyrusOne is a clear top-5 global operator but no longer the #1 independent by pipeline, having been out-raised on equity (Vantage's $9.2B DigitalBridge/Silver Lake round; Blackstone behind QTS). CyrusOne competed by out-levering (CMBS/ABS) rather than out-equitizing — a structurally more fragile way to fund a capex arms race (Lens 13).

Lens 8 · Stock-Price Catalysts → (Overlay) Value-Moving Events

No public stock since 2022. The events that would move an equity value / mark:

1. KKR/GIP take-private, $15B, Mar 2022 — the last "price." Set the $90.50/share reference ``.
2. The 2024 ~$12B financing blitz — validated the CMBS/ABS model and, by implication, a materially higher asset value than 2021 ``.
3. The CME/CyrusOne outage, Nov 26–28 2025 — the defining negative catalyst. A ~10-hour cooling failure at the Aurora, IL (CH2) facility took down CME Globex (gold, US Treasury futures, etc.) for >10 hours; caused by human error (failure to drain cooling towers before a freeze; initial remediation "further exacerbated the problem," cascading chiller failures) . Direct financial consequence: **Goldman paused a ~$1.3B CyrusOne bond sale** . Bloomberg explicitly framed it as "testing the data-center ambitions of KKR and GIP" ``.
4. Powered-land mega-deals (Calpine 400 MW Thad Hill; Constellation 380 MW Freestone; E.ON gas; KEPCO Japan JV) — each a positive value-signal on secured future capacity ``.
5. Community/permitting friction (Whitney/Bosque County water/tax/noise scrutiny) — a slow-drip local-risk catalyst on the US pipeline ``.

Pattern read: for a private infra asset, the value-moving events are (a) the cost/availability of securitized debt and (b) operational reliability — not leasing (leasing is a given in a 1.4%-vacancy market). The Dec-2025 episode showed both risks firing at once.

---

Phase C — Judge people & books

Lens 9 · Management

CEO — Eric Schwartz (since Sep 2022) ``:

• Track record: 16+ years at Equinix, ending as Chief Strategy & Development Officer (corporate strategy, M&A, corporate development, and — tellingly — founder of Equinix's xScale hyperscale JV program) and President of EMEA. This is arguably the single most relevant résumé in the industry for what CyrusOne is now doing: hyperscale JVs + international build-out. A professional operator, not a founder — appropriate for a sponsor-owned scale-up.
• Skin in the game: private — insider ownership n/a — not disclosed. Incentives are almost certainly equity/carry aligned with KKR/GIP's exit, standard for PE-backed CEOs (``, not sourced).
• Bench (strengthened through 2025): John Hatem promoted to President (operations/delivery track record); Krupal Raval joined 2025 as EVP & Chief Strategy Officer — also ex-Equinix xScale MD ``. The pattern is clear: CyrusOne has hired the Equinix xScale playbook wholesale into its C-suite. That is a deliberate, coherent bet on hyperscale-JV execution.

Capital-allocation history: under private ownership the strategy is reinvest everything into powered-land and gigawatt build, funded by securitized debt. No buybacks/dividends (private). The judgment call — aggressive, debt-funded, pro-cyclical expansion into the AI build-out — is high-conviction and, so far, well-timed; but it is the opposite of conservative, and the CMBS-dependence is the vulnerability (Lens 13).

Red flags: the CME outage is a management/operational-integrity red flag, not an accounting one — a "human error / procedures not followed" failure in a mission-critical facility is precisely the kind of controls lapse that should worry a lender and a would-be IPO buyer. Governance transparency is inherently low (private, no public board disclosure). No related-party or promotional-behavior evidence found.

Archetype: professional-manager team executing a sponsor's infra-roll-up thesis — the right archetype for this stage, with the caveat that the whole edifice depends on continued cheap access to asset-backed debt.

Lens 10 · Forensic Red Flags + Regulatory

Accounting risk — mostly not assessable (private, no audited statements since FY2021). Where a public REIT's forensic tells would live (revenue recognition on leases, straight-line rent vs. cash rent, capitalized interest, goodwill from the KKR deal, lease classification, SBC), none is publicly disclosed — n/a — private, not disclosed. This opacity is itself the headline forensic point: an outside analyst cannot verify CyrusOne's cash-flow quality, coverage ratios, or the true leverage behind the $12B/yr issuance. The securitizations provide some third-party discipline (rating agencies, appraised-collateral values — e.g. the $4.4B appraisal on the 2025-1 pool ``), but the consolidated picture is a black box.

Structural forensic concerns (inferred, ``):

• Heavy, structurally-embedded leverage — a securitization-funded build inherently runs high debt/asset; the pro-cyclical CMBS reliance means refinancing risk if AI-capex sentiment or credit spreads turn (the Dec-2025 paused sale is the live example).
• Capitalized interest + development-stage assets — a large in-construction book (~$1B+ per campus) means reported cash yields lag committed capital; classic infra-build timing risk.
• Collateral concentration — securitizations are backed by specific buildings with specific single tenants; a tenant credit event or a facility outage (see CME) directly impairs the collateral.

Regulatory findings (per regulatory/regulatory-findings.md, generated 2026-07-06):

• SEC: "CyrusOne has no CIK — it is private and not required to file with the SEC. No EDGAR enforcement search is possible." 0 SEC Litigation Releases / AAERs ``.
• 10-K Item 3 (Legal Proceedings): no current 10-K exists — n/a — private.
• Non-SEC web search ("CyrusOne" (FTC OR DOJ OR FDA OR CFPB OR settlement OR fine OR penalty)): no material federal enforcement action found . The only live legal/regulatory friction is **local land-use / environmental scrutiny** around the Whitney/Bosque County and Freestone County campuses — water use (aquifer draw, though CyrusOne commits to closed-loop cooling + trucked/well water, *not* Lake Whitney/Brazos River), tax abatements, road-damage remediation, and community noise concerns . No formal lawsuit against CyrusOne surfaced — opposition and negotiated tax-abatement conditions only.
• The CME outage is a potential source of contractual/reputational liability (CME's SLA claims, regulatory attention on critical-market-infrastructure resilience) rather than a securities/enforcement matter ``.

Summary: No material securities-regulatory or enforcement findings — verified via SEC EDGAR EFTS (no CIK → not applicable), web search, and the absence of a public 10-K, as of 2026-07-06. The genuine "red flags" are (1) financial opacity + CMBS-refinancing dependence and (2) the operational-reliability failure at CME — both financial/operational, not forensic-accounting.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection → (Overlay) IPO-Readiness & Path-to-Tradeable

No EPS to project (private, no share count). Under the overlay this lens is (a) a directional financial trajectory and (b) IPO-readiness.

Directional trajectory (all ``, unaudited, wide bands):

• Demand backdrop is as good as it has ever been. Primary-market vacancy hit a record-low ~1.4% at YE2025 (Northern Virginia ~0.3%); wholesale rents up +20–35% YoY, Ashburn breaching $215/kW/month; ~74% of all capacity under construction already pre-leased; pre-leasing timelines out 3–4 years ``. For a landlord with secured power, this is a near-ideal environment.
• Base case: revenue compounds from an estimated ~$2.0–2.5B run-rate at a **high-teens/low-20s % CAGR** as the ~1.16 GW of powered-land + international campuses deliver 2026–2029 — driven by escalators, mark-to-market re-leasing at record rents, and new-capacity lease-up. EBITDA margins on stabilized wholesale assets are structurally **~55–65%**.
• Bull case: the AI build-out sustains the seller's market through 2028; CyrusOne's pre-secured power lets it convert demand faster than power-constrained rivals; a re-IPO or sponsor sale crystallizes a valuation well above the $15B 2022 mark (peers like Vantage raised at scale on far higher implied values).
• Bear case: a credit-spread widening or an AI-capex pause chokes the CMBS/ABS machine; the CME reputational hit raises its cost of debt and slows a sale; single-tenant/facility concentration produces another impairment event.

Forecast tracking: per --watchlist rules, no forecast.ts create is logged here (no committed base case; unattended breadth mode). A future conviction pass could log a binary such as "CyrusOne completes a re-IPO or ≥$20B-implied stake sale by YE2027, p=?".

IPO-readiness (the be-early payoff):

• Stage: late / pre-IPO-capable — a mature, cash-generative, ~$15B+ (2022) platform owned by exit-motivated PE sponsors ~4 years in.
• Readiness (1–5 scale): ~3–4. It has the scale and profitability for an IPO, but two things gate it: (1) the CME outage needs to age out of the narrative before a public listing, and (2) sponsors may prefer a sponsor-to-sponsor or infra-LP stake sale (BlackRock-behind-GIP dynamics; sovereign/infra appetite is high) over a volatile IPO.
• Catalyst: a KKR/GIP monetization event (re-IPO, sponsor sale, or minority stake sale to an infra fund/sovereign) is the single event that makes CyrusOne tradeable — plausible window 2026–2028.
• Write-back note: CyrusOne is absent from research/private-watch.json despite being a top-tier private datacenter name — it should be added (suggested: beat: datacenters, stage: late, ipo_readiness: 3, lead_investors: "KKR, GIP/BlackRock", catalyst: "sponsor exit / re-IPO window 2026–2028; CME-outage overhang", dossier: this file). Not written in this unattended run — flagged for the master session (wave boundary: no edits to research state files here).

Lens 12 · Bull vs Bear

Bull case. CyrusOne owns three of the scarcest things in the AI economy: secured gigawatt-scale power, delivery track record at hyperscale, and top-tier sponsor capital — in the tightest commercial-real-estate market on record (1.4% vacancy, rents +20–35%). It pioneered data-center securitization, giving it a repeatable ~$12B/yr funding engine, and it has hired the Equinix xScale brain-trust to run an international hyperscale-JV playbook (Japan/KEPCO, Italy). Every incremental megawatt it delivers leases at record economics on 10–15 yr contracts. As power — not capital — became the binding constraint, its early powered-land deals (Calpine/Constellation/E.ON) became a moat competitors physically cannot replicate in the timeframe. A 2026–2028 sponsor exit likely crystallizes a value well north of the $15B 2022 take-private.

Bear case. CyrusOne is a levered bond-machine wearing a growth-company costume. Three things could permanently impair it: (1) the securitization flywheel is pro-cyclical and spread-sensitive — the Dec-2025 paused $1.3B Goldman bond sale proved the funding can seize exactly when it's needed; a credit shock or AI-capex pause turns "growth capex" into "refinancing wall." (2) Revenue is concentrated in a handful of hyperscalers, one tenant per building — the same buyers who are its customers are also its most dangerous competitors (they self-build) and its counterparty risk. (3) The CME outage exposed operational fragility in mission-critical facilities — a "human error" chiller cascade that took down global markets is a reputational and contractual-liability event that raises its cost of capital and complicates an exit. And it has been out-scaled on the wholesale-development leaderboard by Vantage and QTS, who out-equitized it — CyrusOne chose the more fragile, debt-funded path.

Pre-mortem (18 months out, thesis broke): Most likely story — a 2026 credit-spread widening (or an AI-capex "digestion" scare) froze the data-center ABS/CMBS market; CyrusOne, mid-build on multiple gigawatt campuses with a large capitalized-interest book, faced a refinancing wall; the CME-outage overhang had already widened its spreads; KKR/GIP were forced to inject equity or accept a down-round stake sale, and the "$20B+ exit" narrative evaporated. The demand was never the problem — the financing structure was.

Are multiples too high? Unknowable directly (private). But the sector is priced for a permanent seller's market; the contrarian risk is that 74% pre-leasing + 3–4 yr pipelines are pulling forward supply that normalizes vacancy — and rents — by 2028–2029, right as CyrusOne's biggest capacity delivers.

Contrarian view (what the market refuses to see): The consensus obsesses over demand (AI is infinite, vacancy is 1%, rents only go up). The thing being ignored: for a securitization-funded operator, the binding risk is the liability side, not the asset side. CyrusOne's fate is decided in the CMBS market and in its chiller rooms — not in its leasing pipeline. The Dec-2025 fortnight (an outage and a paused bond sale, same week) was the market showing its hand; most observers filed it under "one-off." It wasn't — it was the shape of the real risk.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• The moat is a commodity in a cyclical costume. A powered wholesale shell has no IP, no network effect, no ecosystem lock-in. The "moat" is secured power + capital + delivery speed — all of which Vantage (bigger pipeline, $9.2B equity), QTS (Blackstone), Aligned, STACK, and the hyperscalers' own self-build teams also have. In wholesale, the landlord competes on price and speed; margins compress the moment supply catches demand.
• Revenue concentration is structural and severe. One tenant per building; a short list of hyperscalers; the same customers are the most dangerous competitors (Microsoft/Meta/AWS self-build at massive scale and can insource). If two or three anchor tenants slow leasing or renegotiate at renewal, the securitized collateral behind billions in CMBS is impaired.
• The most dangerous competitor bulls underrate: the customer itself. Every hyperscaler is building its own capacity; the "buy vs. build" line moves against colos when hyperscalers have the balance sheet (they do) and the power queue position. CyrusOne is a swing supplier for demand its customers can't self-serve fast enough — a good place to be in a boom, a terrible place in a digestion year.
• The capital structure is the kill-switch. ~$12B raised in 2024 entirely as debt — warehouse facilities, CMBS, ABS. This works only while the securitization market is open and spreads are tight. It is the single most spread-sensitive, pro-cyclical funding model in infrastructure. The Dec-2025 Goldman pause is not an anecdote — it's a stress test the company failed in miniature. In a 2008- or 2022-style credit event, a large in-construction book funded by rolling asset-backed paper is exactly what breaks.
• Worst capital-allocation risk: doubling down on debt-funded, pro-cyclical gigawatt build at the top of an AI-capex cycle, with governance opacity that prevents outsiders from seeing the true coverage ratios.
• What must hold for today's (implied) value: (1) the ABS/CMBS market stays open at tight spreads through the full build; (2) hyperscaler demand doesn't digest before 2028–2029; (3) no second CME-class operational event; (4) sponsors find a rich exit before any of the above breaks.
• If growth disappoints 20–30% (leasing slows / a digestion year): stabilized-asset value holds, but the in-construction book's value and the refinancing terms deteriorate sharply — the equity (most junior in a levered structure) takes the hit, and the exit multiple compresses.
• Single scenario that permanently impairs the business: a credit-market freeze coinciding with an AI-capex pause, forcing distressed refinancing / a sponsor down-round while multiple gigawatt campuses are mid-build. Plausibility: moderate — not a base case in a 1.4%-vacancy world, but the Dec-2025 episode moved it from "tail" to "live."

Lens 14 · Management Questions (ordered by information value)

1. What is consolidated total debt / total assets, and net-debt / stabilized-EBITDA, today — and how much of the debt matures or must be refinanced in the next 24 months? (The single number that decides the whole thesis.)
2. After the December 2025 Goldman bond-sale pause, what were your realized funding spreads on the next issuance versus the prior one — i.e., what did the CME outage cost you in cost of capital?
3. What share of contracted revenue comes from your top 3 tenants, and what is the weighted-average remaining lease term (WALT)?
4. What is the stabilized vs. in-construction split of your asset base, and what is the total capitalized-interest / development-in-progress balance?
5. What concrete operational-controls changes followed the CME/Aurora cooling failure, and what contractual liability (SLA credits, claims) resulted?
6. Of your ~1.16 GW+ of secured powered-land, how much is contracted to a paying tenant versus speculative/available?
7. How do you underwrite the buy-vs-build risk — at what point do your hyperscaler customers insource, and what's your defense?
8. What is the exit plan and timeline for KKR/GIP — re-IPO, sponsor sale, or infra-LP stake sale — and what milestones gate it?
9. What are your renewal mark-to-market assumptions, and what happens to them if primary-market vacancy normalizes from 1.4% toward 5%+ by 2028–2029?
10. What is the development yield-on-cost you're underwriting for new campuses, and how has it moved with construction/power/equipment inflation?
11. How exposed are you to long-lead electrical-gear (transformers/switchgear) and what's your delivery-schedule risk from that?
12. How does the KEPCO Japan JV and European gas-behind-the-meter strategy pencil on returns versus your US powered-land model?
13. What is your customer pipeline concentration by AI/neocloud vs. hyperscaler vs. enterprise, and how do you weigh neocloud counterparty (credit) risk?
14. How do you think about local permitting/water/tax friction (Whitney, Freestone) as a constraint on the US pipeline?
15. What would have to be true for you to slow the build — what's your trigger to stop deploying capital pro-cyclically?

---